Daily Briefing

What Changed

The deep learning model outperformed LAA‑950 in segmentation accuracy and correlated better with radiologist scores and pulmonary function tests. [1]

Medical Imaging | Deep learning segmentation beats LAA‑950 on CT emphysema quantification. Journal article: In a study using virtual data and multi‑center clinical cohorts—including 101 patients in cohort C1 and 23 in C2—the deep‑learning model achieved higher Dice scores than LAA‑950 (virtual 76.6%±8.8 vs 51.5%±23.5; C1 48.4%±24.2 vs 22.5%±20.2; C2 64.8%±9.3 vs 32.7%±16.8) and showed reduced bias and limits of agreement in C1 (1.0%±4.3 vs 2.1%±15.1) with stronger correlations to Fleischner scores (0.77 vs 0.47) and DLCO (multivariate R² 0.31‑0.32 vs 0.21‑0.25) in cohort C3. [1]

Research & Evidence | LLM after-visit summaries scored higher on understandability and actionability than physician versions. Journal article: In a blinded retrospective study of 50 discharged adults, GPT-4 and Gemma summaries achieved higher PEMAT understandability (85.5%, 87.5% vs 66.1%) and actionability (70.9%, 74.1% vs 56.7%; all p<.001) and were rated better on the AVSrubric for clarity, comprehensibility, empathy and lower potential for harm. [2]

Industry & Products | AI‑driven tool tracks physician tasks per contract in Toledo pilot. Podcast from Healthcare AI Pioneers: A speaker described an AI‑driven product that lets physician leaders log tasks aligned with their contracts, originating in the Toledo market with input from physicians and the compliance team who conceptualized it, and helps decrease the risk associated with administering those contracts. [3]

Industry & Products | Oura files for IPO with $1.2B nine‑month revenue and 5M members. Oura’s S‑1 shows $1.2B revenue for the nine months ended June 30, up 74% YoY, and $61M net income on 55% gross margins. It has 5M paid members wearing the ring 23 h/day, 94% activation‑to‑membership conversion and 85% annual retention, plus 42B hours of biometric data and 1,200+ integration partners. [4]

Clinical Practice | Speaker used AI to retrieve U.S. ALS organization contacts in about fifteen minutes. Podcast from The Business of AI in Healthcare: A speaker said they used AI to get names, phone numbers and details of all U.S. ALS organizations, a task that would have taken weeks manually, in about fifteen minutes. [5]